Initial Fedora 32 vs. Fedora 33 Beta Benchmarks Point To Slightly Higher Performance

[oleid]

I use Fedora 32 and Btrfs really Looks Like a Sapontage in the Performance

Btrfs should Not be the Default

The filesystem change seems to bring some performance regressions... I'll keep ext4 for a while so.

For god's sake, disable copy on write for the database or the directory containing the database.

chattr +C /dir/file

Everybody using btrfs for databases knows that. And it should be default in the benchmark as well.

[ferry]

I use Fedora 32 and Btrfs really Looks Like a Sapontage in the Performance

Btrfs should Not be the Default

Because you will notice a sqlite slowdown, eh when?

While a useful functionality change like snapshots, you won't be using knowingly?

Do you use a spellchecker in your browser, or does it slow down the performance too much?

[birdie]

The geometric mean with the ~200 tests actually points to Fedora Workstation 33 coming out slightly ahead of Fedora Workstation 32 -- both out-of-the-box and with the stable release updates that positions it closer to the F33 versions.

Michael

The last chart with the geometric mean of all results indicates that F33 is the slowest among them all. Weird.

[JackLilhammers]

See the linked OpenBenchmarking.org page in the review for all the data points, it wasn't practical including all ~200 benchmarks especially for the ones where there is virtually no change.

You're right, my bad. Thank you

[oleid]

Michael

The last chart with the geometric mean of all results indicates that F33 is the slowest among them all. Weird.

Really? Not in my browser. It is indicated that _more is better_.

[AdamOne]

Michael, is there any reason to do F33 on EXT4 comparisons?

[sunnyflunk]

See the linked OpenBenchmarking.org page in the review for all the data points, it wasn't practical including all ~200 benchmarks especially for the ones where there is virtually no change.

My BS meter was going haywire (especially after reviewing all benchmarks) so I exported the data, removed the tests where 3 results weren't present and standardized each test where the lowest result =1 and then averaged.

My results were:

F32 Workstation 1.0379
F32 Workstation Updated 1.0364
F33 Workstation Beta 1.0156

So F33 was 2% slower and the 32 ones were basically the same...

# Tests where result was more than 5% better than worst result

F32 Workstation 23
F32 Workstation Updated 25
F33 Workstation Beta 12

[Guest]

Largely unrelated, but to anyone running or who could quickly access a F33 beta install, does xrandr allow for the --filter option? I specifically want to use that option to get some crude HiDPI scaling on a game.

As I understand, at some point in F32, xrandr didn't support that option (I think it was outdated but not sure). Apparently Ubuntu 20.04's xrandr doesn't support that option. openSUSE TW's xrandr supports the option.

[skeevy420]

Largely unrelated, but to anyone running or who could quickly access a F33 beta install, does xrandr allow for the --filter option? I specifically want to use that option to get some crude HiDPI scaling on a game.

As I understand, at some point in F32, xrandr didn't support that option (I think it was outdated but not sure). Apparently Ubuntu 20.04's xrandr doesn't support that option. openSUSE TW's xrandr supports the option.

Likely output for you from F33 Silverblue. Hope this helps.

$ xrandr --filter
xrandr: unrecognized option '--filter'
Try 'xrandr --help' for more information.
$ xrandr --help
usage: xrandr [options]
where options are:
--display <display> or -d <display>
--help
-o <normal,inverted,left,right,0,1,2,3>
or --orientation <normal,inverted,left,right,0,1,2,3>
-q or --query
-s <size>/<width>x<height> or --size <size>/<width>x<height>
-r <rate> or --rate <rate> or --refresh <rate>
-v or --version
-x (reflect in x)
-y (reflect in y)
--screen <screen>
--verbose
--current
--dryrun
--nograb
--prop or --properties
--fb <width>x<height>
--fbmm <width>x<height>
--dpi <dpi>/<output>
--output <output>
--auto
--mode <mode>
--preferred
--pos <x>x<y>
--rate <rate> or --refresh <rate>
--reflect normal,x,y,xy
--rotate normal,inverted,left,right
--left-of <output>
--right-of <output>
--above <output>
--below <output>
--same-as <output>
--set <property> <value>
--scale <x>x<y>
--scale-from <w>x<h>
--transform <a>,<b>,<c>,<d>,<e>,<f>,<g>,<h>,<i>
--off
--crtc <crtc>
--panning <w>x<h>[+<x>+<y>[/<track:w>x<h>+<x>+<y>[/<border:l>/<t>/<r>/<b>]]]
--gamma <r>:<g>:<b>
--brightness <value>
--primary
--noprimary
--newmode <name> <clock MHz>
<hdisp> <hsync-start> <hsync-end> <htotal>
<vdisp> <vsync-start> <vsync-end> <vtotal>
[flags...]
Valid flags: +HSync -HSync +VSync -VSync
+CSync -CSync CSync Interlace DoubleScan
--rmmode <name>
--addmode <output> <name>
--delmode <output> <name>
--listproviders
--setprovideroutputsource <prov-xid> <source-xid>
--setprovideroffloadsink <prov-xid> <sink-xid>
--listmonitors
--listactivemonitors
--setmonitor <name> {auto|<w>/<mmw>x<h>/<mmh>+<x>+<y>} {none|<output>,<output>,...}
--delmonitor <name>
[rialpa84@localhost ~]$

[sunnyflunk]

F32 Workstation 1.0379
F32 Workstation Updated 1.0364
F33 Workstation Beta 1.0156

Actually, should update the numbers, LIB results were using 2 - 1/(x) where x is how many times slower the worst result was. So for the sqlite result which is 3.33x slower on F33, it was only giving a result of 1.7 on F32. So LIB was capped at 2 (with a zero result!) while HIB is uncapped.

Mean
F32 Workstation 1.0537
F32 Workstation Updated 1.0512
F33 Workstation Beta 1.0185

Or geometric mean:
F32 Workstation 1.0430
F32 Workstation Updated 1.0414
F33 Workstation Beta 1.0170